Efficient Knowledge Distillation of SAM for Medical Image Segmentation

Ganapathy Krishnamurthi; Gowthamaan Palani; Kunal Dasharath Patil

REVIEW 5 major objections 5 minor 26 references

Efficient Knowledge Distillation of SAM for Medical Image Segmentation

T0 review · 5 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims that KD SAM, a 26.4M-parameter distillation of SAM trained with MSE plus perceptual loss and decoder fine-tuning, reaches comparable or better Dice than SAM and MobileSAM on most of the four medical datasets.

desk verdict A reasonable engineering variant of MobileSAM whose own Table 1 undercuts the 'comparable or superior' claim. read the letter →

arxiv 2501.16740 v1 pith:B5ODT37E submitted 2025-01-28 eess.IV cs.AIcs.CV

Kunal Dasharath Patil , Gowthamaan Palani , Ganapathy Krishnamurthi This is my paper · ORCID

classification eess.IVcs.AIcs.CV

keywords knowledgedistillationSegmentAnythingModelmedicalimagesegmentationperceptuallossencoderdecoderfine-tuningDiceResNet-50

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

KD SAM is the paper's proposal for shrinking the Segment Anything Model for medical image segmentation: it replaces SAM's 632M-parameter ViT-H encoder with a modified ResNet-50 (26.4M total parameters) and then fine-tunes the SAM decoder. The training is decoupled -- the encoder is distilled first using the sum of mean squared error and a perceptual loss from a pre-trained feature extractor, and the decoder is fine-tuned afterwards with Dice loss while the encoder stays frozen. The paper reports Dice scores of 0.9774 on Fetal Head Ultrasound, 0.9114 on ISIC 2017, 0.8586 on Kvasir-SEG, and 0.8216 on Breast Ultrasound, and it summarizes these as comparable or superior to SAM and MobileSAM on most of these tasks. If the claim holds, interactive prompt-driven segmentation at SAM-like accuracy could run on hardware with roughly one twenty-fourth of SAM's parameter budget.

What carries the argument

The machinery that carries the argument is the decoupled distillation pipeline behind KD SAM: a combined loss $L_{\mathrm{Combined}} = L_{\mathrm{MSE}} + L_{\mathcal{P}}$ on encoder feature maps, where $L_{\mathrm{MSE}}$ is pixel-wise mean squared error between teacher and student feature maps and $L_{\mathcal{P}}$ is a perceptual loss computed from selected layers of a pre-trained convolutional feature extractor; the student encoder is a ResNet-50 with channels trimmed to 256 and upsampling layers to match the teacher's spatial resolution. After encoder distillation, the SAM decoder is fine-tuned with Dice loss while the encoder is frozen, so the decoder adapts to the student's feature distribution.

What would settle it

Run SAM, MobileSAM, and KD SAM on the same held-out test folds with identical, fixed prompts (e.g., ground-truth bounding boxes for every image) and compare Dice; the central claim is falsified if KD SAM's mean Dice is not within the paper's reported margins on Fetal Head and ISIC 2017, or if its deficits on Kvasir-SEG and Breast Ultrasound are larger than reported.

Watch

Extended reading notes

Core claim

The central claim is that a decoupled distillation can preserve SAM's medical segmentation quality at a fraction of its size. The paper trains a modified ResNet-50 encoder to imitate SAM's ViT-H feature maps under the combined loss $L_{\mathrm{Combined}} = L_{\mathrm{MSE}} + L_{\mathcal{P}}$, where $L_{\mathrm{MSE}}$ is the mean squared error between teacher and student feature maps and $L_{\mathcal{P}}$ is a perceptual loss from selected pre-trained feature-extractor layers, then fine-tunes the SAM mask decoder with Dice loss while the encoder is frozen. The resulting 26.4M-parameter KD SAM is reported to reach Dice scores of 0.9774 on Fetal Head Ultrasound and 0.9114 on ISIC 2017, above both baselines, while scoring 0.8586 on Kvasir-SEG and 0.8216 on Breast Ultrasound, below one or both baselines. The paper's summary claim is that KD SAM is comparable or superior to SAM and MobileSAM across most of the four datasets.

Load-bearing premise

The comparison assumes that SAM and MobileSAM were evaluated under the same prompt protocol -- same type and quality of point or box prompts -- as KD SAM, but the paper never reports what prompts the baselines received.

Editorial extensions

If this is right

At 26.4M parameters, KD SAM is roughly 24x smaller than SAM, so a faithful reproduction of the reported results would put prompt-driven medical segmentation in reach of mobile and edge hardware.
The reported Fetal Head (0.9774) and ISIC 2017 (0.9114) Dice scores exceed both SAM and MobileSAM, indicating that on these two tasks the distilled student does not pay an accuracy penalty.
The decoupled recipe -- encoder distillation with MSE plus perceptual loss, followed by frozen-encoder decoder fine-tuning with Dice loss -- is the reusable component that other distillation efforts could adopt.
The paper's own numbers show the tradeoff is not uniform: Kvasir-SEG is 0.8586 versus 0.8715 for SAM, and Breast Ultrasound is 0.8216 versus 0.9051, so the claim of comparable or superior performance rests on how the per-dataset results are aggregated.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The paper never specifies the prompts used to obtain the SAM and MobileSAM baselines, so the cross-model ranking in Table 1 is provisional until the evaluation protocol is pinned down; this is an editorial caution, not a result in the paper.
The same two-stage loss could be ablated by training the student with MSE only, perceptual only, and the sum, to see which term drives the gains on fetal-head and melanoma boundaries; the paper presents only the combined loss.
Because the drop is largest on Breast Ultrasound, a modality with heterogeneous echo patterns, the distillation may transfer unevenly across imaging domains; per-modality calibration of the perceptual loss layers is a natural next experiment.
With MobileSAM at 5M parameters and KD SAM at 26.4M, there is a wide middle range of the accuracy-efficiency curve; one could sweep student backbones to locate the Pareto frontier, which the paper does not explore.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Quick take: this is a modest, clearly written engineering paper that applies MobileSAM's decoupled distillation to a ResNet-50 student and adds a perceptual loss. The headline claim of 'comparable or superior' performance does not survive contact with Table 1: on Breast Ultrasound KD SAM scores 0.8216 Dice versus 0.9051 for SAM and 0.8985 for MobileSAM, an 8-point gap; on Kvasir it is 0.8586 versus 0.8715 and 0.8719. On half the datasets the student is clearly worse, and on the other two the gains are marginal. The abstract and Section 5 overstate what the table shows.

What is new is limited but real: the specific combination of a modified ResNet-50 student (channels reduced to 256) with unweighted MSE plus perceptual loss for encoder distillation, evaluated on four medical datasets, is not in the cited papers. The decoupled two-phase training is practical and clearly described, and the related work is honestly cited, including the paper's own admission that the recipe is 'as done by MobileSAM'. The qualitative figures provide some visual support.

The soft spots are mostly in the evaluation. No prompting protocol is given for SAM or MobileSAM, and SAM's accuracy is highly prompt-dependent. Without knowing whether the baselines used a center point, a bounding box, or multiple clicks, the numbers in Table 1 are not well-defined. There are no error bars, seeds, or statistical tests, and the train/test split is not described. The decoder is fine-tuned per dataset and evaluated on held-out data, which is fine in principle, but the details are missing. Equation (3) is internally inconsistent: the text says perceptual loss uses VGG features, but the equation defines phi_l^T and phi_l^S as feature maps from the ViT and ResNet encoders. Also, the 'significantly fewer parameters' claim is only true against SAM (632M); MobileSAM has 5M, so KD SAM at 26.4M is not the smallest lightweight option.

Who is this for? A reader tracking lightweight SAM variants for medical segmentation might want a peek, but the results as presented do not establish an advantage over MobileSAM. The paper deserves a serious referee, though major revisions are needed: report the prompt protocol, add error bars and reproducibility details, fix Eq. (3), and rewrite the claim in the abstract. I would not cite this as a reliable comparison until those issues are addressed.

Referee Report

5 major / 5 minor

Summary. The paper proposes KD SAM, a two-phase knowledge-distillation approach for compressing the Segment Anything Model (SAM) for medical image segmentation. In the first phase, a ResNet-50 encoder is trained to mimic SAM's ViT-H encoder using a combination of mean squared error and perceptual loss. In the second phase, the SAM decoder is fine-tuned on medical segmentation datasets with Dice loss while the distilled encoder is frozen. The authors evaluate KD SAM on Kvasir-SEG, ISIC 2017, Fetal Head Ultrasound, and Breast Ultrasound, reporting Dice coefficients in Table 1 and claiming comparable or superior performance to SAM and MobileSAM with significantly fewer parameters.

Significance. If the claimed performance held, a 26.4M-parameter medical image segmenter approaching SAM's accuracy would be practically valuable for resource-constrained settings. The paper also addresses a real problem, since SAM and MobileSAM have limitations in medical segmentation. However, the manuscript's central claim is not supported by its own reported numbers, and the experimental protocol is too incomplete to establish the validity of the comparison. The method itself is plausible and builds on existing distillation ideas, but the evidence as presented is not sufficient to justify the conclusions.

major comments (5)

[Abstract and Table 1] The central claim that KD SAM achieves 'comparable or superior performance to the baseline models' is contradicted by the numbers in Table 1 on two of four datasets. On Breast Ultrasound, KD SAM obtains 0.8216 versus 0.9051 for SAM and 0.8985 for MobileSAM, an approximately 8-point gap. On Kvasir-SEG, KD SAM scores 0.8586 versus 0.8715 and 0.8719. The abstract and Section 5 therefore overstate the results; the claim is only partially supported (Fetal Head, ISIC) and even there the differences are within a range that requires statistical validation.
[Section 4 and Table 1] The evaluation protocol for the SAM and MobileSAM baselines is not specified. SAM is a prompt-based interactive model whose accuracy depends strongly on the prompt type (point, box, mask) and prompt quality. The paper never states what prompts were used for the baseline models, how many prompts, or whether ground-truth boxes were used. Without this information, the baseline Dice scores in Table 1 are not well-defined and the comparison cannot be reproduced or trusted. This is a load-bearing omission because the entire conclusion hinges on a fair baseline comparison.
[Section 5, parameter comparison] The statement that KD SAM has 'significantly fewer parameters' than the baseline models is misleading. The paper reports 26.4M parameters for KD SAM versus 632M for SAM, which is indeed far fewer, but MobileSAM has only 5M parameters. Thus KD SAM has substantially more parameters than MobileSAM. The sentence in Section 5 acknowledges this but the abstract does not. The efficiency claim should be qualified to compare against SAM only, or should explain why 26.4M is still advantageous relative to MobileSAM in the medical imaging context.
[Section 3.1, Eq. (3)] The definition of perceptual loss in Eq. (3) is internally inconsistent. The text states that perceptual loss uses a pre-trained VGG network, but Eq. (3) defines φ^T_l and φ^S_l as feature maps from 'the ViT and ResNet encoders, respectively.' The notation also applies C_l, H_l, W_l, which are spatial/depth dimensions suited to convolutional feature maps, but ViT outputs are sequences of token embeddings. As written, the loss cannot be implemented without additional clarification of how VGG is used and how the feature maps are aligned.
[Sections 3.2 and 4] The inference protocol for KD SAM itself is not described. The paper says the decoder is fine-tuned with Dice loss, but it does not state whether prompts are used at test time and, if so, what kind and how many. For an interactive segmentation model, the test-time prompting protocol is essential for interpreting the reported Dice scores. Without this, the reader cannot determine what the numbers in Table 1 mean, even for the proposed method.

minor comments (5)

[Section 4] The paper does not specify train/validation/test splits, dataset preprocessing details, or the random seeds used for training and evaluation. These details are necessary for reproducibility.
[Section 4] The modification of ResNet-50 (reducing channels from 2048 to 256 and adding upsampling layers) is described only briefly; the exact architecture and how the spatial dimensions are matched to the ViT feature maps should be given explicitly.
[Section 5] The sentence 'ResNet-50 as several considerations drove the student model' in Section 3.1 is ungrammatical; it should be rephrased, e.g., 'Several considerations drove the selection of ResNet-50 as the student model.'
[Section 5] Figures 2 and 3 are referenced but not explained in the text; the reader is not told what the box plot and bar chart show or how the qualitative examples in Figure 3 were selected.
[Section 2] The related work discussion would benefit from a direct comparison with MobileSAM's decoupled distillation approach, since the proposed method appears similar in spirit; the paper should clarify the novel contribution relative to MobileSAM.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: KD SAM is an empirical distillation study whose results are held-out comparisons, not reductions of its own inputs.

full rationale

This paper is an empirical knowledge-distillation study, not a derivation from first principles. The teacher (SAM ViT-H) is an external published model; the student encoder is trained with standard MSE and perceptual losses, and the decoder is fine-tuned with Dice loss. The central claim rests on the Dice-score comparison in Table 1, and the paper explicitly says the model 'was evaluated on a separate test dataset.' No parameter is fitted to the reported Dice values and then renamed as a prediction; no load-bearing step is justified by a self-citation (the reference list contains no self-citations); no uniqueness theorem is invoked; and the ResNet-50 encoder is not defined in terms of the evaluation metric. The ambiguity about the prompt protocol for SAM/MobileSAM baselines and the partially inaccurate 'fewer parameters' statement are correctness and reproducibility concerns, not circular-reasoning defects. The derivation chain is therefore self-contained, and the appropriate circularity score is 0.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on hand-chosen loss weights, an unspecified architecture modification, and an undisclosed evaluation protocol. No new physical or architectural entities are introduced beyond a standard modified ResNet-50.

free parameters (3)

MSE/Perceptual loss weighting = 1:1 (unweighted sum in Eq. 1)
No ablation or justification; central to how features are distilled.
ResNet-50 output channel size and upsampling configuration = 2048 reduced to 256 with upsampling layers (unspecified count)
Hand-chosen architecture modification to match teacher feature dimensions; not ablated.
Perceptual loss layer selection = unspecified 'selected layers' of VGG
Eq. 3 sums over layers l but the specific layers are not given, so the loss is not reproducible.

assumptions (4)

domain assumption SAM ViT-H is a suitable teacher for medical segmentation
The whole method assumes teacher features contain transferable medical-image knowledge; no specific medical-domain evidence is provided.
domain assumption Adding perceptual loss improves distilled features over MSE alone
Motivated by related work on EdgeSAM; no ablation in this paper.
domain assumption Standard train/test splits of the four datasets are used for decoder fine-tuning and evaluation
The paper says 'separate test dataset' but does not specify the split, so the evaluation relies on this assumption.
domain assumption Dice loss is sufficient to align the decoder with the distilled encoder
No comparison to other decoder losses or joint fine-tuning is provided.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Efficient Knowledge Distillation of SAM for Medical Image Segmentation." pith.science (2026). https://pith.science/paper/B5ODT37E

@misc{pith2026250116740,
  author       = {Pith},
  title        = {Pith review of: Efficient Knowledge Distillation of SAM for Medical Image Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/B5ODT37E}},
  note         = {Machine review of arXiv:2501.16740}
}

read the original abstract

The Segment Anything Model (SAM) has set a new standard in interactive image segmentation, offering robust performance across various tasks. However, its significant computational requirements limit its deployment in real-time or resource-constrained environments. To address these challenges, we propose a novel knowledge distillation approach, KD SAM, which incorporates both encoder and decoder optimization through a combination of Mean Squared Error (MSE) and Perceptual Loss. This dual-loss framework captures structural and semantic features, enabling the student model to maintain high segmentation accuracy while reducing computational complexity. Based on the model evaluation on datasets, including Kvasir-SEG, ISIC 2017, Fetal Head Ultrasound, and Breast Ultrasound, we demonstrate that KD SAM achieves comparable or superior performance to the baseline models, with significantly fewer parameters. KD SAM effectively balances segmentation accuracy and computational efficiency, making it well-suited for real-time medical image segmentation applications in resource-constrained environments.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 17 canonical work pages

[1]

The Segment Any- thing Model (SAM) [1] has established itself as a powerful tool in this domain, leveraging a Vision Transformer (ViT)

INTRODUCTION Interactive image segmentation has become a cornerstone in numerous applications, including medical imaging, au- tonomous driving, and augmented reality. The Segment Any- thing Model (SAM) [1] has established itself as a powerful tool in this domain, leveraging a Vision Transformer (ViT)

work page
[2]

However, the significant computational demands of SAM hinder its deployment in real-time and resource-constrained environ- ments, such as mobile devices and edge platforms

encoder and prompt-guided mask decoder to achieve high segmentation accuracy across diverse datasets. However, the significant computational demands of SAM hinder its deployment in real-time and resource-constrained environ- ments, such as mobile devices and edge platforms. MobileSAM [3] addresses these limitations by replacing the ViT encoder with ViT-Ti...

work page
[3]

Initially de- veloped for classification tasks [5], it has been adapted for dense prediction tasks such as semantic segmentation [6] and object detection [7]

RELA TED WORK Knowledge distillation is a powerful technique to transfer knowledge from a large, complex model to a smaller model while maintaining considerable performance. Initially de- veloped for classification tasks [5], it has been adapted for dense prediction tasks such as semantic segmentation [6] and object detection [7]. The goal is to align the...

work page
[4]

EXPERIMENTAL SETUP The training process for the knowledge distillation framework was carried out on multiple medical imaging datasets, includ- ing Kvasir-SEG [14], ISIC 2017 [15], Fetal Head Ultrasound [16], and Breast Ultrasound [17]. These datasets were chosen for their diversity and relevance to the segmentation tasks, providing a comprehensive evaluat...

work page 2017
[5]

This approach addresses the computational limitations of SAM’s Vision Transformer (ViT) encoder by distilling its knowledge into a lightweight ResNet [12] based encoder

METHOD The proposed method adapts the Segment Anything Model (SAM) for medical image segmentation through a tailored decoupled knowledge distillation process as done by Mobile- SAM. This approach addresses the computational limitations of SAM’s Vision Transformer (ViT) encoder by distilling its knowledge into a lightweight ResNet [12] based encoder. As sh...

work page
[6]

Structured knowledge distil- lation for semantic segmentation,

Yifan Liu, Ke Chen, Chris Liu, Zengchang Qin, Zhenbo Luo, and Jingdong Wang, “Structured knowledge distil- lation for semantic segmentation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 2604–2613

work page 2019
[7]

Learning efficient object de- tection models with knowledge distillation,

Guobin Chen, Wongun Choi, Xiang Yu, Tony Han, and Manmohan Chandraker, “Learning efficient object de- tection models with knowledge distillation,” Advances in neural information processing systems, vol. 30, 2017

work page 2017
[8]

As shown in Table 1, the results demonstrate that KD SAM achieves comparable or superior performance to the base- line models across most datasets

RESULTS The performance of the KD SAM model was evaluated on a separate test dataset using the Dice Coefficient metric across four medical imaging datasets: Kvasir-SEG, ISIC 2017, Fetal Head Ultrasound, and Breast Ultrasound, and com- pared against the baseline models, SAM and MobileSAM. As shown in Table 1, the results demonstrate that KD SAM achieves co...

work page 2017

Show all 26 references

[9]

However, this comes at the cost of segmenta- tion quality, as it diverges from SAM’s interactive segmen- tation principles, making it less suitable for high-precision tasks

instance segmentation model combined with heuristic post-processing rules for object selection to achieve faster segmentation. However, this comes at the cost of segmenta- tion quality, as it diverges from SAM’s interactive segmen- tation principles, making it less suitable fo...

2025 arXiv
[10]

Segment anything,

Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C Berg, Wan-Yen Lo, et al., “Segment anything,” in Proceedings of the IEEE/CVF International Conference on Computer Vi- sion, 2023, pp. 4015–4026

2023
[11]

An image is worth 16x16 words: Transformers for image recognition at scale,

Alexey Dosovitskiy, “An image is worth 16x16 words: Transformers for image recognition at scale,” arXiv preprint arXiv:2010.11929, 2020

2010 arXiv
[12]

Chaoning Zhang, Dongshen Han, Yu Qiao, Jung Uk Kim, Sung-Ho Bae, Seungkyu Lee, and Choong Seon Hong, “Faster segment anything: Towards lightweight (a) Kvasir-SEG (b) Fetal Head (c) ISIC 2017 (d) Breast Ultrasound Fig. 3 . Comparison of Segmentation Results against SAM and Mobi...

2017 arXiv
[13]

networks, to capture high-level semantic similarities be- tween the two models. This loss function evaluates the dis- tance between the feature representations of the teacher and student models at multiple layers, ensuring that the student model not only replicates the low-lev...
[14]

Per- ceptual losses for real-time style transfer and super- resolution,

Justin Johnson, Alexandre Alahi, and Li Fei-Fei, “Per- ceptual losses for real-time style transfer and super- resolution,” in Computer Vision–ECCV 2016: 14th Eu- ropean Conference, Amsterdam, The Netherlands, Oc- tober 11-14, 2016, Proceedings, Part II 14 . Springer, 2016, pp. 694–711

2016
[15]

Differ- entiable feature aggregation search for knowledge dis- tillation,

Yushuo Guan, Pengyu Zhao, Bingxuan Wang, Yuanxing Zhang, Cong Yao, Kaigui Bian, and Jian Tang, “Differ- entiable feature aggregation search for knowledge dis- tillation,” in Computer Vision–ECCV 2020: 16th Euro- pean Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Pa...

2020
[16]

Fast segment anything,

Xu Zhao, Wenchao Ding, Yongqi An, Yinglong Du, Tao Yu, Min Li, Ming Tang, and Jinqiao Wang, “Fast segment anything,” arXiv preprint arXiv:2306.12156 , 2023

2023 arXiv
[17]

Yolact: Real-time instance segmentation,

Daniel Bolya, Chong Zhou, Fanyi Xiao, and Yong Jae Lee, “Yolact: Real-time instance segmentation,” in Pro- ceedings of the IEEE/CVF international conference on computer vision, 2019, pp. 9157–9166

2019
[18]

Efficientsam: Leveraged masked image pretraining for efficient seg- ment anything,

Yunyang Xiong, Bala Varadarajan, Lemeng Wu, Xiaoyu Xiang, Fanyi Xiao, Chenchen Zhu, Xiaoliang Dai, Dilin Wang, Fei Sun, Forrest Iandola, et al., “Efficientsam: Leveraged masked image pretraining for efficient seg- ment anything,” in Proceedings of the IEEE/CVF Con- ference on ...

2024
[19]

Edgesam: Prompt-in-the-loop distillation for on-device deployment of sam,

Chong Zhou, Xiangtai Li, Chen Change Loy, and Bo Dai, “Edgesam: Prompt-in-the-loop distillation for on-device deployment of sam,” arXiv preprint arXiv:2312.06660, 2023

2023 arXiv
[20]

Deep residual learning for image recognition,

Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun, “Deep residual learning for image recognition,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 770–778

2016
[21]

Very deep convolutional networks for large-scale image recogni- tion,

Karen Simonyan and Andrew Zisserman, “Very deep convolutional networks for large-scale image recogni- tion,” arXiv preprint arXiv:1409.1556, 2014

2014 arXiv
[22]

Kvasir-seg: A segmented polyp dataset,

Debesh Jha, Pia H Smedsrud, Michael A Riegler, P ˚al Halvorsen, Thomas De Lange, Dag Johansen, and H˚avard D Johansen, “Kvasir-seg: A segmented polyp dataset,” in MultiMedia modeling: 26th international conference, MMM 2020, Daejeon, South Korea, Jan- uary 5–8, 2020, proceedin...

2020
[23]

Analysis of the isic image datasets: Usage, benchmarks and recommendations,

Bill Cassidy, Connah Kendrick, Andrzej Brodzicki, Joanna Jaworek-Korjakowska, and Moi Hoon Yap, “Analysis of the isic image datasets: Usage, benchmarks and recommendations,” Medical image analysis , vol. 75, pp. 102305, 2022

2022
[24]

Automated mea- surement of fetal head circumference using 2d ultra- sound images,

Thomas LA van den Heuvel, Dagmar de Bruijn, Chris L de Korte, and Bram van Ginneken, “Automated mea- surement of fetal head circumference using 2d ultra- sound images,” PloS one, vol. 13, no. 8, pp. e0200412, 2018

2018
[25]

Dataset of breast ultrasound images,

Walid Al-Dhabyani, Mohammed Gomaa, Hussien Khaled, and Aly Fahmy, “Dataset of breast ultrasound images,” Data in brief, vol. 28, pp. 104863, 2020

2020
[26]

Adam: A method for stochastic optimization,

Diederik P Kingma, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980, 2014

2014 arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.

[1] [1]

The Segment Any- thing Model (SAM) [1] has established itself as a powerful tool in this domain, leveraging a Vision Transformer (ViT)

INTRODUCTION Interactive image segmentation has become a cornerstone in numerous applications, including medical imaging, au- tonomous driving, and augmented reality. The Segment Any- thing Model (SAM) [1] has established itself as a powerful tool in this domain, leveraging a Vision Transformer (ViT)

work page

[2] [2]

However, the significant computational demands of SAM hinder its deployment in real-time and resource-constrained environ- ments, such as mobile devices and edge platforms

encoder and prompt-guided mask decoder to achieve high segmentation accuracy across diverse datasets. However, the significant computational demands of SAM hinder its deployment in real-time and resource-constrained environ- ments, such as mobile devices and edge platforms. MobileSAM [3] addresses these limitations by replacing the ViT encoder with ViT-Ti...

work page

[3] [3]

Initially de- veloped for classification tasks [5], it has been adapted for dense prediction tasks such as semantic segmentation [6] and object detection [7]

RELA TED WORK Knowledge distillation is a powerful technique to transfer knowledge from a large, complex model to a smaller model while maintaining considerable performance. Initially de- veloped for classification tasks [5], it has been adapted for dense prediction tasks such as semantic segmentation [6] and object detection [7]. The goal is to align the...

work page

[4] [4]

EXPERIMENTAL SETUP The training process for the knowledge distillation framework was carried out on multiple medical imaging datasets, includ- ing Kvasir-SEG [14], ISIC 2017 [15], Fetal Head Ultrasound [16], and Breast Ultrasound [17]. These datasets were chosen for their diversity and relevance to the segmentation tasks, providing a comprehensive evaluat...

work page 2017

[5] [5]

This approach addresses the computational limitations of SAM’s Vision Transformer (ViT) encoder by distilling its knowledge into a lightweight ResNet [12] based encoder

METHOD The proposed method adapts the Segment Anything Model (SAM) for medical image segmentation through a tailored decoupled knowledge distillation process as done by Mobile- SAM. This approach addresses the computational limitations of SAM’s Vision Transformer (ViT) encoder by distilling its knowledge into a lightweight ResNet [12] based encoder. As sh...

work page

[6] [6]

Structured knowledge distil- lation for semantic segmentation,

Yifan Liu, Ke Chen, Chris Liu, Zengchang Qin, Zhenbo Luo, and Jingdong Wang, “Structured knowledge distil- lation for semantic segmentation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 2604–2613

work page 2019

[7] [7]

Learning efficient object de- tection models with knowledge distillation,

Guobin Chen, Wongun Choi, Xiang Yu, Tony Han, and Manmohan Chandraker, “Learning efficient object de- tection models with knowledge distillation,” Advances in neural information processing systems, vol. 30, 2017

work page 2017

[8] [8]

As shown in Table 1, the results demonstrate that KD SAM achieves comparable or superior performance to the base- line models across most datasets

RESULTS The performance of the KD SAM model was evaluated on a separate test dataset using the Dice Coefficient metric across four medical imaging datasets: Kvasir-SEG, ISIC 2017, Fetal Head Ultrasound, and Breast Ultrasound, and com- pared against the baseline models, SAM and MobileSAM. As shown in Table 1, the results demonstrate that KD SAM achieves co...

work page 2017